AI_DLC-nano

AI-DLC Nano - ship fixes and features, not paperwork

An AI-driven development workflow for personal and medium-sized projects that keeps just enough written state to be pausable, resumable, and auditable - and spends everything else on building and testing code.

The problem

AI coding agents are great at writing code and terrible at continuity. Close the session and the context is gone. Switch from Claude Code to Cursor and you start over. Full-weight AI-DLC processes solve this with formal requirements docs, user stories, and design documents - overkill when the “team” is you and the project is a side project.

The answer

Five tiny documents, six lean phases, four platforms.

Document	Purpose	Size
tech-stack.md	Stack, conventions, test commands	< 40 lines
codebase-map.md	Structural orientation (brownfield, only if non-trivial)	30–50 lines
intent.md (per work item)	What, why, and every confirmed decision	< 1 screen
plan.md (per work item)	Task checklist + test approach	< 1 screen
state.md + audit.md	Where you are, and how you got there	lines, not pages

Phases: Bootstrap → Intake → Clarify → Plan → Construct → Wrap-up. You approve each step forward, and four gates are never skipped: generating code, choosing the branch, any action that leaves your machine (tracker/Slack/push/PR/deploy), and destructive local actions. Prefer speed? Say “autopilot” and the between-phase check-ins drop away - the four hard gates stay.

Why people stay

• Resume anywhere. State lives in your repo. Pause on the laptop with Claude Code tonight, resume on the desktop with Cursor tomorrow - the agent reads state.md and continues mid-checklist.
• Decisions, remembered. The AI asks one batched round of rationale questions, records the answers, and never asks again. Six months later, audit.md tells you why the code is the way it is.
• Tracker-aware, human-gated. Pulls GitHub issues or Jira tickets automatically when an MCP is connected; posts summaries back only after you approve the exact text.
• Drop-in. Copy one folder (.claude/, .cursor/, .codex/, or .github/) into any repo - greenfield or brownfield - and type /ai-dlc-nano.
• Token-thrifty. Every artifact has a size cap. Your budget goes to construction, not ceremony.

Get started in 60 seconds

Inspired by AWS AI-DLC. Stripped to what a personal project actually needs.